1
OpenAI 应用研究主管谈空间推理，称 Astra 模型已补上人类少数优势
IT之家 9 月 6 日消息，OpenAI 应用研究主管鲍里斯 · 鲍尔（Boris Power）今天在 X 平台表示， 空间推理曾是人类大幅领先计算机的少数优势之一 ，如今有了 Astra 模型，这一差距已不复存在。 据悉，独立 AI 研究员 @spicylemonade 曾在今天早些时候表示，A…
2
LEGO-Puzzles: How Good Are MLLMs at Multi-Step Spatial Reasoning?
乐高拼图当考题，多模态大模型的多步空间推理能力一测便知，机器人、自动驾驶都靠它。
arXiv:2503.19990v4 Announce Type: replace Abstract: Many real-world applications of spatial intelligence, such as robotic control, autonomous driving,…
3
Disentangling 3D Modeling from Spatial Reasoning
这项研究将3D建模与空间推理解耦，为AI生成更精细的3D场景提供新思路。
arXiv:2608.05242v1 Announce Type: new Abstract: In this work, we explore an alternative paradigm for spatial reasoning by explicitly disentangling 3D …
4
Trace, Verify, and Correct: A Training-Free Framework for Spatial Reasoning in Multimodal LLMs
无需训练，三步纠错，多模态大模型空间推理能力再进一步
arXiv:2608.04759v1 Announce Type: new Abstract: Although Multimodal Large Language Models (MLLMs) have made substantial progress, their spatial reason…
5
Ask HN: How do you correct spatial reasoning of LLMs?
大模型的空间推理短板如何补救？HN 上工程师晒出 3D 打印实战难题，围观各路高手支招。
I'm working on a complex multi-part 3D-printed product, and I needed some engineering input from LLMs. I use highest Gemini Pro reasoning levels, Fabl…
6
SymboUQ: Symbolic Uncertainty Quantification for Spatial Reasoning in LLMs
大模型空间推理常“答非所证”，这项研究用符号不确定性量化给结论可靠性加把锁，值得关注。
arXiv:2608.00417v1 Announce Type: new Abstract: Although large language models (LLMs) can produce fluent spatial reasoning traces, their intermediate …
7
Explainable and Resource-Efficient Spatial Reasoning in Multimodal LLMs for Decision-Critical Applications
多模态大模型的空间推理可解释性与资源效率新方法，助力机器人等决策关键场景提升信任与审计能力
arXiv:2607.27145v1 Announce Type: new Abstract: As Multimodal Large Language Models (MLLMs) are increasingly deployed in decision-critical pipelines s…
8
Beyond Medical Diagnostics: How Medical Multimodal Large Language Models Think in Space
医疗多模态大模型突破图像诊断边界，首次系统揭示其空间思维机制
arXiv:2603.13800v2 Announce Type: replace Abstract: Visual spatial intelligence is critical for medical image interpretation, yet remains largely unex…
9
Towards Enhancing 3D Spatial Reasoning in Medical Multimodal Large Language Models
新方法突破医学大模型空间推理瓶颈，让AI看懂CT三维结构，提升诊断精准度
arXiv:2607.13860v1 Announce Type: new Abstract: While Multimodal Large Language Models (MLLMs) have demonstrated remarkable success in 2D medical imag…
10
Multimodal Continuous Reasoning via Asymmetric Mutual Variational Learning
突破语言空间瓶颈，用非对称互变分学习解锁多模态连续推理新路径
arXiv:2607.00461v1 Announce Type: new Abstract: Multimodal Large Language Models (MLLMs) are often constrained by a language-space bottleneck, forcing…
11
SpatialMosaic: A Multiview VLM Dataset for Partial Visibility
新数据集SpatialMosaic专攻部分可见场景，补全多视图VLM空间推理短板。
arXiv:2512.23365v4 Announce Type: replace Abstract: Recent progress in Multimodal Large Language Models (MLLMs) has enabled 3D scene understanding and…
12
Dense Reward for Multi-View 3D Reasoning with Global Maps and Local Views
用全局地图搭配局部视图，为多视角3D推理提供稠密奖励信号，突破空间理解瓶颈。
arXiv:2606.23557v1 Announce Type: new Abstract: Multi-view 3D Visual Question Answering (MV3D-VQA) requires integrating partial observations into a co…
13
EndoCoT: Scaling Endogenous Chain-of-Thought Reasoning in Diffusion Models
探索扩散模型中内生思维链推理新机制，突破MLLM文本编码器的推理瓶颈。
arXiv:2603.12252v4 Announce Type: replace-cross Abstract: Recently, Multimodal Large Language Models (MLLMs) have been widely integrated into diffusio…
14
OneCanvas: 3D Scene Understanding via Panoramic Reprojection
用全景画布统一多视图特征，无需复杂几何编码器，实现轻量级3D场景理解。
arXiv:2606.19253v1 Announce Type: cross Abstract: Existing approaches to 3D scene understanding in Vision-Language Models (VLMs) either rely on comple…
15
Ouroboros-Spatial: Closing the Data-Model Loop for Spatial Reasoning
提出数据-模型闭环框架Ouroboros-Spatial，动态生成训练样本，自适应提升多模态大模型空间推理能力，突破静态数据集局限。
arXiv:2606.11719v1 Announce Type: cross Abstract: Spatial reasoning remains a persistent challenge for multimodal large language models (MLLMs). Exist…
16
Language-based Trial and Error Falls Behind in the Era of Experience
大语言模型在语言任务中称王，但面对符号、空间等非语言环境却表现不佳，研究揭示其本质局限
arXiv:2601.21754v3 Announce Type: replace Abstract: While Large Language Models (LLMs) excel in language-based agentic tasks, their applicability to u…
17
Sample-Efficient Post-Training for LEGO Spatial-Physics Reasoning
提出样本高效后训练方法，突破乐高空间物理推理任务的数据瓶颈
arXiv:2606.07602v1 Announce Type: new Abstract: LLM-based LEGO assembly generation requires both semantic grounding and physical feasibility. We ident…
18
Skill-3D: Evolving Scene-Aware Skills for Agentic 3D Spatial Reasoning
揭示现有方法在3D场景中工具误用与偏好偏差，提出场景感知技能学习新范式
arXiv:2606.07436v1 Announce Type: new Abstract: This paper explores agentic 3D spatial understanding, i.e., MLLM agents performing 3D reasoning throug…
19
Thinking with Imagination: Agentic Visual Spatial Reasoning with World Simulators
VLMs空间推理受限于观察图像和文本思维链，新方法用世界模拟器实现想象推理，突破跨视角一致性与未观测布局推断难题。
arXiv:2606.06476v1 Announce Type: new Abstract: While Vision-Language Models (VLMs) have shown strong visual reasoning capabilities, their spatial rea…
20
Eliciting Complex Spatial Reasoning in MLLMs through Wide-Baseline Matching
CVPR 2026 新作：用宽基线匹配技术激发多模态大模型的空间推理潜力，突破复杂场景理解瓶颈。
arXiv:2606.03577v1 Announce Type: new Abstract: Wide-baseline matching (WBM) requires integrating geometric understanding, viewpoint changes, fine-gra…